Amazon is reportedly acquiring a large number of rare books, scanning their pages for AI training purposes, and destroying the books in the process. This action aligns Amazon with other major tech firms like Anthropic and Meta, which are also engaged in extensive book scanning and disposal activities.
The aim, as noted in some reports, is to feed AI systems high-quality writing instead of the, let’s say, less-than-stellar material often found on the internet. To cut costs and expedite the process, bookbinding machines are being used to sever the spines of books, according to a comprehensive report detailing operations in unmarked warehouses affiliated with Amazon.
A news outlet managed to uncover Amazon’s book procurement practices by embedding tracking devices into rare books they believed would be acquired. They accurately traced these books’ journeys to an Amazon facility in Las Vegas.
An Amazon representative stated, “Amazon purchases books through commercial channels to help us develop and improve the products and services our customers use.”
The unidentified bookseller cooperated with the investigation by including Apple AirTags in a batch of around 1,000 books supplied through the Book Marketplace in July.
Warehouse workers in Las Vegas have indicated that their daily tasks involve receiving numerous printed books and prepping them for scanning. This intricate process reportedly makes use of automated machines to eliminate bindings from orphaned books, echoing what was previously covered in a January piece from a major publication focused on book-scanning efforts by various AI entities.
Interestingly, the mascot for the Amazon team in the warehouse is a T. rex named VGT3, which is depicted as munching on a book. One employee commented on an online forum about their role at VGT3, saying, “I work here at VGT3 in Las Vegas, and all we do is scan books.” They mentioned that some colleagues are responsible for cutting books while others manage the scanning of barcodes.
Another worker reiterated, “Working at VGT3 is fun because all I do is scan books,” expressing how enjoyable the job is.
According to the reported findings, some bookstores have seen a notable increase in sales over the past year, likely due to the high demand for training materials from AI companies.
While there is some uncertainty, booksellers suspect that the AI giant is behind the surge in bulk orders; it appears they are purchasing a peculiar number of randomly selected books, often without concern for pricing, unlike libraries or academic institutions.
Books hold particular importance for AI Labs, especially those published before 2022, as they are assured not to include AI-generated content. Companies assert that training models using AI-generated texts can lead to a troubling phenomenon known as “model collapse.”
An Anthropic co-founder expressed that exposing AI to well-written books could ultimately enhance writing quality rather than mimicking the subpar communication found online. An internal Meta email from early 2024 emphasized the necessity of safeguarding the digital repository of books to stay competitive in the AI landscape.
Reports indicate that around early 2024, a project called “Project Panama” was initiated by Humanity executives to secretly scan virtually every book available.
According to a document released in a legal filing, Anthropic’s internal plan stated, “Project Panama is our initiative to destructively scan every book in the world. We don’t want people to know we’re working on this.”
Initial consideration was given by Anthropic to acquire books from used bookstores and libraries, including the New York Public Library, which has been described as having chronic funding issues.
